I keep tabs open when I want to come back to something. But it leads to me having 300+ tabs open so Ill put things I want to do here instead of clogging up my computer memory :P
To do:
- write nanochat copy from scratch
- write nanoGPT from scratch
- Watch all of Andrejs youtube channel
- Watching entire umar jamil youtube channel
- Finish the vidoe gen post
- Derive GAN loss and make sure it is right for video (diffusion gan)
- Derive the SiD loss and place it in the framework
- make demo of log space smoothing
To read:
- TTT done right
- https://arxiv.org/html/2510.23972v1
- https://lilianweng.github.io/posts/2024-04-12-diffusion-video/
- all of https://lilianweng.github.io/
- https://kris.pengy.ca/designforlearning
- TRPO paper
- Paul Grahm
- Score Identity Distillation
- Diffusion models and th emanifold hypothesis Log-Domain smoothing is Geometry Adaptive
- Equilibrium Matching: Generative Modeling with Implicit ENergy-Based Models
To watch:
- https://www.youtube.com/watch?v=dRuhl6MLC78
- https://www.youtube.com/watch?v=CBYhVcO4WgI&list=PLU630Cd0ZQCMeQiSvU7DJmDJDitdE7m7r&index=1
GAN
classes grading
